AI Expo Korea 2026: Kaytus Launches All-QLC Flash Storage for 10,000-GPU Clusters

Key Takeaways

• All‑QLC solution delivers 10 TB/s aggregate bandwidth for 10k‑GPU clusters
• 100 million random‑read IOPS achieved at exabyte scale
• Five‑year TCO cut by 70 % versus TLC‑based storage
• Power consumption reduced over 75 % with QQLC flash
• KR1180 offers 1 PB capacity in a 1U chassis

Pulse Analysis

The surge in generative‑AI workloads has exposed a fundamental mismatch between traditional storage architectures and the read‑intensive demands of massive model training. Enterprises typically shuffle petabytes of data from object stores into parallel file systems, incurring latency penalties and consuming up to 81 % of data‑team effort on preparation. Moreover, TLC flash, optimized for write endurance, inflates capital and operational expenditures without delivering tangible benefits for predominantly read‑heavy AI pipelines. This storage gap threatens GPU utilization rates, forcing data centers to over‑provision compute to compensate for I/O starvation.

Kaytos’s all‑QLC flash platform tackles these pain points through a hardware‑software co‑design that pairs PCIe 5.0 direct‑connect servers with AI‑native parallel file systems and GPU Direct Storage. By unifying file, object, and block access under a single namespace, the solution eliminates cross‑system data migration, delivering sub‑millisecond latency and fully linear scaling from petabyte to exabyte capacities. Benchmark results—10 TB/s sustained bandwidth and 100 million IOPS—show that even a 10,000‑GPU cluster can maintain over 95 % GPU utilization, a dramatic improvement over legacy tiered storage that suffers from metadata lock contention.

Beyond performance, the economic impact is compelling. The QLC‑based design reduces five‑year TCO by 70 % and cuts power and cooling costs by more than 75 %, translating into multi‑million‑dollar savings for hyperscale AI providers. As AI model sizes continue to grow, cost‑effective, high‑throughput storage will become a decisive factor in platform selection, positioning Kaytus as a potential disruptor in the enterprise AI infrastructure market.